PASS Summit 2013 Keynote, Day 2

Today is the second day of the 2013 SQL PASS Summit, and I’m again live blogging the event.  image

8:15: Here we go.  Looks like a thinner crowd today.  Everyone sleeping in from #sqlkaraoke last night?

8:19: The morning is kicked off with a video montage about networking.  It’s good to see emphasis on getting to know your peers.

8:21: Douglas McDowell talking about PASS finance.  A full 75% of PASS revenue comes from the annual PASS Summit.  Looks like the new Business Analytics conference contributes significantly (about 20%) to the budget, generating about 100k in net profit.  Interesting that SQL Rally is not mentioned.  Interesting note that PASS has been working on building up a rainy day fund, and the organization now has $1MM in financial reserves.  Good financial update – relatively brief and to the point.

8:30: Bill Graziano recognizes the outgoing board members, Douglas McDowell and Rushahb Mehta.

8:35: Tom LaRock, new PASS president, takes the stage.  He announces that PASSTv has reached 3,000 people in 79 countries this year.  He also recognizes the incoming ExecCo and board members for the PASS board.  He announces that the PASS Business Analytics Conference (BAC) will be held in San Jose, CA in May of next year, and the next PASS Summit will return to Seattle in 2014.

8:41: Dr. David DeWitt takes the stage.  He’s always a crowd favorite.  Heads may explode in the next hour.  He’ll be talking about Hekaton, the in-memory database technology.  He starts off by poking some fun at the marketing team and their habitual renaming of products.

8:45: Hekaton is memory-optimized but durable.  If the power goes out, you haven’t lost anything.  It is fully baked into SQL Server 2014, which was released in CTP2 earlier this week. The aim for Hekaton is 100x performance improvement, which cannot be gained through improvements elsewhere (CPU, etc.).  Hekaton is more than just pinning tables in memory, which in itself would not yield the expected performance gains.  Hekaton is an engine, not a bolt-on.

8:56: DeWitt talks about database concurrency, showing a small example of competing write operations.  Concurrency, locks, and serialization, oh my.  I think I just heard the first head explosion of the morning.  The slide deck can be downloaded here

9:05: Hekaton uses a timestamp mechanism rather than latching.  Related: lock-free data structures.  DeWitt confesses that lock-free data structures are really hard to understand, but gives a couple of brief examples (via animation) of how performance of latches versus lock-free structures operates under a workload.  Hekaton concurrency control is based on three techniques: optimistic data read/write, multiversioning, and timestamping.  When he describes the timestamp mechanism with start/end valid periods, end timestamp of infinity used to represent the current row, it sounds a little like an in-memory version of a slowly changing dimension.  A non-blocking garbage collector will clean up expired or no-longer-valid rows as a transparent background process.

9:37: I have to admit that this stuff blows my mind.  As a BI practitioner, I pay attention to database performance and concurrency issues, but I’ve never dived into the engine in this level of detail.  Though I’m not a performance guru, it’s apparent to me that this stuff is truly revolutionary.  Note to self to find a way to experiment with this soon.

9:57: David DeWitt wraps up.  Great stuff, and it’s obvious the crowd was behind him the whole way (even if he blew minds along the way).

Bookmark the permalink.